Re: [PATCH V3] Expose resource control capabilites on cache bank

[Eli Qiao <qiaoliyong@xxxxxxxxx>]

On Thursday, 6 April 2017 at 9:04 PM, Martin Kletzander wrote:

On Thu, Apr 06, 2017 at 01:25:35PM +0100, Daniel P. Berrange wrote:

On Thu, Apr 06, 2017 at 08:20:56PM +0800, Eli Qiao wrote:

This patch is based on Martin's cache branch.

This patch amends the cache bank capability as follow:

<cache>

<bank id='0' level='3' type='unified' size='15360' unit='KiB' cpus='0-5'>

<control min='768' unit='KiB' type='unified' nallocations='4'/>

</bank>

Why do we need to report 'type' on both bank & control elements. Are they

really expected to have different values ?

There’s a discussion from https://www.redhat.com/archives/libvir-list/2017-March/msg01689.html

I think I made a mistake here, it should be ’scope’ instead of ’type’ here.

No matter what’s it should be, at least we need to identify if the host enabled CDP for L3 cache or not.

If enabled CDP on host, the resctrl looks like this:

root@s2600wt:/sys/fs/resctrl# cat schemata

L3DATA:0=fffff;1=fffff

L3CODE:0=fffff;1=fffff

root@s2600wt:/sys/fs/resctrl# cat info/L3

L3CODE/ L3DATA/

root@s2600wt:/sys/fs/resctrl# cat info/L3DATA/num_closids

8

root@s2600wt:/sys/fs/resctrl# cat info/L3CODE/num_closids

8

as you can see that num_closids are divided into 2 part (compared to 16 without CDP),

 but other info are still same

root@s2600wt:/sys/fs/resctrl# cat info/L3CODE/cbm_mask

fffff

root@s2600wt:/sys/fs/resctrl# cat info/L3CODE/min_cbm_bits

1

root@s2600wt:/sys/fs/resctrl# cat info/L3DATA/cbm_mask

fffff

root@s2600wt:/sys/fs/resctrl# cat info/L3DATA/min_cbm_bits

1

------------------------------------------

below output are from disable CDP:

root@s2600wt:/sys/fs/resctrl# cat schemata

L3:0=fffff;1=fffff

root@s2600wt:/sys/fs/resctrl# cat info/L3/

cbm_mask      min_cbm_bits  num_closids

root@s2600wt:/sys/fs/resctrl# cat info/L3/cbm_mask

fffff

root@s2600wt:/sys/fs/resctrl# cat info/L3/min_cbm_bits

1

root@s2600wt:/sys/fs/resctrl# cat info/L3/num_closids

16

 

That's one of my questions I had way back in some of the previous

discussions. Did not get the answer. I suspect there is a reason to

whether CDP is enabled or not and it is not just the type of the cache

bank itself. If it is, then CDP makes no sense for us at all, actually

for anyone who has access to cache information and that would mean bad

design of that thing. And that's not something I would expect from this

functionality.

Regards,

Daniel

--

|: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :|

|: http://libvirt.org -o- http://virt-manager.org :|

|: http://entangle-photo.org -o- http://search.cpan.org/~danberr/ :|

--

libvir-list mailing list

libvir-list@xxxxxxxxxx

https://www.redhat.com/mailman/listinfo/libvir-list

--
libvir-list mailing list
libvir-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/libvir-list